TOPLINE

In patients with heart failure (HF) with reduced ejection fraction and New York Heart Association (NYHA) functional class III, a smartwatch-based AI system using continuous heart rate monitoring accurately predicted rises in levels of N‑terminal pro-B‑type natriuretic peptide (NT-proBNP), provided alerts earlier than weight-based monitoring, according to a brief report published in JACC: Heart Failure.

METHODOLOGY

• Researchers conducted a multicenter prospective observational study (TRIBE-HF II) between January 2022 and May 2024 to evaluate whether a passive smartwatch‑based AI platform (CardioID) could predict worsening HF.
• The study included adults aged 18-85 years with symptomatic HF with ejection fraction < 40% and NYHA functional class III. The derivation cohort included 91 patients, and the validation cohort included 33.
• Patients wore a Sony mSafety smartwatch paired with the CardioID smartphone app that recorded their heart rate. Baseline and follow-up blood samples every 14 days collected using home kits were used to measure NT-proBNP levels.
• After 2 weeks of calibration, the algorithm tracked 14 days’ data on heart rate to raise alerts for NT-proBNP events — either a > 600 pg/mL absolute rise or a more than 100% relative rise — within a 35-day window, indicating worsening HF.
• The primary endpoint was the diagnostic performance of the algorithm for detecting NT-proBNP events compared with virtual weight‑gain alerts, which triggered when weight increased > 3 lb in 1 day or > 5 lb in 3 days. Both were set to two alerts per patient‑year. The median follow-up duration was 182 days.

TAKEAWAY

• In the validation cohort, the CardioID algorithm achieved an area under the curve (AUC) of 0.865. It flagged 82.6% of NT‑proBNP events and correctly gave no alert for 89.2% without increases in NT‑proBNP levels; 51.4% of alerts matched true rises; and correctly did not send alerts 97.4% of the time.
• At the matched alert rate, CardioID detected cases of true rises in NT‑proBNP levels more often than weight-based monitoring (40.8% vs 7.2%), alerts matched true rises more often (60.0% vs 37.5%), and non-alerts were correct more often (92.0% vs 69.9%; P ≤ .05 for all). The AUC did not differ significantly between the two approaches.
• Among true positive alerts, 89.5% of CardioID alerts occurred 2 days before the NT‑proBNP event and 73.7% occurred at least 7 days before. CardioID alerted earlier than weight‑based monitoring (median alert lead time, 14 days vs 4 days).
• Patients had higher adherence to CardioID than to weight‑based monitoring (100% vs 60.6%; P < .001).

IN PRACTICE

“CardioID demonstrated strong diagnostic performance with greater sensitivity, adherence, and longer alert lead time than weight-based monitoring, supporting its ability to detect physiologic deterioration earlier,” the researchers wrote.

SOURCE

The study was led by Abhishek Chaturvedi, MD, of Centre for Chronic Disease Control in New Delhi, India. It was published online on June 30 as a brief report in JACC: Heart Failure.

LIMITATIONS

The study had a single-arm observational design and only included patients with NYHA functional class III. The size of the validation cohort was small. Researchers used NT-proBNP events instead of adjudicated HF hospitalizations.

DISCLOSURES

The study was sponsored by General Prognostics, and two authors reported being co-founders of the sponsoring company. One author reported receiving research support from and serving as a consultant for several pharmaceutical and healthcare companies including General Prognostics. Another author reported receiving institutional grants from government agencies.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.